To solve the problems of unstable and low reliability calibration results due to the random selection of calibrated pose points, the number of optimal pose point and the selection algorithm of optimal pose set based on the singular value of Jacobian matrix to calculate observable indexes are studied. The MDH model is formulated and the Levenberg-Marquardt (LM) algorithm is used to identify geometric parameters. The points of the Staubli TX60 robot end-effector selected based on the optimal and random pose set are measured by the LeicaAT960 laser tracker. On the basis of analyzing and studying the uncertainty contributors of robot calibration, the GUM method is used to calculate the uncertainty of geometric parameter calibration and Monte Carlo simulation method is utilized to evaluate the uncertainty of robot end-effector pose, respectively. Results show that the accuracy of the robot calibrated by the optimal pose set is not only greatly improved at the test points, but also the mean uncertainty of geometric parameters and end-effector is about 0. 11 times of that calibrated by the random pose set. The calibration results are stable and reliable, and the generalization ability is strong, which are suitable for popularization and application in high-precision and large-scale operation situations.